LISREL output files for measurement models

1. Introduction
In practice, the variables of interest are often latent (unobservable) variables, such as
intelligence, job satisfaction, organizational commitment, socio-economic status, ambition,
alienation, verbal ability, etc. These latent variables are modeled by specifying a
measurement model and a structural model. The measurement model specifies the
relationships between the observed indicators and the latent variables while the structural
equation model specifies the relationships amongst the latent variables.

In practice, it is very convenient to refine the measurement model before the complete
model is fitted to the data. In this regard, the reliability and the validity of the indicators of
the latent variables are assessed even before the measurement model is fitted to the data.
Thereafter, modification indices are sometimes used to refine the measurement model.

LISREL 8.80 for Windows (Jöreskog & Sörbom 2006) does not include methods to assess
the reliability (internal consistency) and validity of the indicators of a measurement model.
However, it does include the Maximum Likelihood (ML), Robust Maximum Likelihood
(RML), Generalized Least Squares (GLS), Un-weighted Least Squares (ULS), Weighted
Least Squares (WLS), Diagonally Weighted Least Squares (DWLS) and Full Information
Maximum Likelihood (FIML) methods to fit measurement models to data. Each of these
methods includes modification indices that may be used to refine measurement models.
More information on these indices is provided in (Jöreskog & Sörbom 1999) and Du Toit &
Du Toit (2001). In addition, LISREL 8.80 for Windows include methods to assess the
validity of latent constructs as outlined in Steenkamp & Van Trijp (1991).

In this note, the RML method of LISREL 8.80 for Windows is used to fit and refine a
measurement model to the scores of a sample of school children on nine psychological
tests. This data set is described in the next section. The measurement model is described
in section 3. In section 4, SIMPLIS syntax file for the measurement model is described
while the results are presented in section 5. The modification indices are then used to
refine the model in section 6.

LISREL output files for measurement models 1

2. The data
The data are the scores on nine psychological tests of 145 seventh- and eighth-grade
students at the Grant-White school in Chicago. The raw data are listed in the PRELIS
System File (PSF) NPV.PSF located in C:\LISREL 8.8 Student Examples\WORKSHOP.
The first couple of lines of this file are shown in the following PSF window.

LISREL output files for measurement models 2

3. The measurement model
We consider nine psychological tests that were administered to 145 seventh- and eighth-
grade students. These nine psychological tests are theoretically constructed to measure
the visual perception, verbal ability and numerical speed of seventh- and eighth-grade
children. A path diagram for the corresponding measurement model for visual perception,
verbal ability and numerical speed is shown in Figure 1.

Figure 1: A path diagram for a measurement model of Visual Perception,

Verbal Ability, and Numerical Speed

LISREL output files for measurement models 3

4. Fitting the measurement model to the data
Generating the estimated Asymptotic Covariance Matrix (ACM)
¾ Use the Open option on the File menu to load the Open dialog box.
¾ Browse for the location C:\LISREL 8.8 Student Examples\WORKSHOP.
¾ Select the PRELIS Data (*.psf) option from the Files of type dropdown list box.
¾ Select the file NPV.PSF by clicking on it.
¾ Click on the Open button to open the PSF window for NPV.PSF.
¾ Select the Output Options option on the Statistics menu to load the Output dialog
box.
¾ Check the Save to File checkbox in the Asymptotic Covariance Matrix section.
¾ Enter the name NPV.ACM in the string box in the Asymptotic Covariance Matrix
section.
¾ Click on the OK button to run PRELIS28 to generate the text editor window containing
the output file NPV.OUT.

The SIMPLIS syntax file

¾ Use the Open option on the File menu to load the Open dialog box.
¾ Browse for the location C:\LISREL 8.8 Student Examples\WORKSHOP.
¾ Select the file NPV.SPL by clicking on it.
¾ Click on the Open button to open the following text editor window.

Line 1 provides a title for the analysis.

Line 2 specifies the raw data source.
Line 3 specifies the binary file containing the estimated asymptotic matrix of the sample
variances and covariances.
Lines 4 and 5 specify labels for the latent variables.
Lines 6 thru 9 specify the measurement model to be fitted to the data.
Line 10 specifies that the results should be printed in terms of the LISREL model for the
measurement model in Figure 1. It also requests three decimal places, the completely
standardized solution and the modification indices.
Line 11 requests a path diagram of the fitted model.
Line 12 indicates no more SIMPLIS commands are to be processed.

LISREL output files for measurement models 4

¾ Click on the Run LISREL icon on the main toolbar to produce the PTH window for
NPV.PTH.

5. The results
The output file, NPV.OUT, consists of several sections. In this section, we will review some
selections of the output file. The sample covariance matrix is shown in the following text
editor window.

The parameter specifications for the LISREL model for the model in Figure 1 are listed in
the following text editor window.

LISREL output files for measurement models 5

The estimated matrix of factor loadings follows.

LISREL output files for measurement models 6

The results above indicate that all the free factor loadings are statistically significant if a
significance level of 1% is used. Consequently, the construct validity of the nine
psychological tests as indicators of the three latent variables appears to be at an
acceptable level.

The estimated correlation matrix of the three latent variables is provided in the following
text editor window.

LISREL output files for measurement models 7

The results above indicate that all three factor correlations are statistically significant if a
significance level of 1% is used. In other words, there is sufficient evidence that the three
abilities are correlated.

The estimated measurement error variances are shown below.

The results above indicate that all the measurement error variances are statistically
significant if a significance level of 1% is used.

The estimated squared multiple correlations of the nine psychological tests are shown in
the following text editor window.

The results above indicate that the latent variable verbal ability explain more of the variation
in all three its indicators than the latent variables visual and numerical ability.

LISREL output files for measurement models 8

The measures of fit of the measurement model are displayed in the following two text editor
windows.

LISREL output files for measurement models 9

These measures of fit indicate that the model provides a poor fit to the data.

The next two text editor windows show the modification indices for the matrix of factor
loadings and the measurement error covariance matrix.

LISREL output files for measurement models 10

The results above indicate that the largest modification index is that for the factor loading of
visual ability for the indicator S-C CAPS.

LISREL output files for measurement models 11

The completely standardized solution is given in the following two text editor windows.

The completely standardized estimates above indicate that the solution is admissible since
none of them exceeds unity in absolute value.

LISREL output files for measurement models 12

6. Refining the measurement model
From the results in section 5, it is evident that the largest modification index is that of the
factor loading of the variable S-C CAPS as an indicator of latent variable Visual. The
contents of this test support this suggestion. In other words, this factor loading makes
sense from a theoretical point of view. We will now demonstrate how to modify the
SIMPLIS syntax file for the measurement model to accommodate this factor loading.
The SIMPLIS syntax file
¾ Use the Open option on the File menu to load the Open dialog box.
¾ Browse for the location C:\LISREL 8.8 Student Examples\WORKSHOP.
¾ Select the file NPV1.SPL by clicking on it.
¾ Click on the Open button to open the following text editor window.

Line 1 provides a title for the analysis.

Line 2 specifies the data source.
Lines 3 to 4 specify labels for the latent variables.
Lines 5 to 9 specify the measurement model to be fitted to the data.
Line 10 specifies that the results should be printed in terms of the LISREL model for the
measurement model in Figure 1. It also requests three decimal places, the completely
standardized solution and modification indices.
Line 11 requests a path diagram.
Line 12 indicates no more SIMPLIS commands are to be processed.

¾ Click on the Run LISREL icon on the main toolbar to produce the following PTH
window.

LISREL output files for measurement models 13

The modification indices are displayed in the following two text editor windows.

LISREL output files for measurement models 14

From the results above, it is evident that none of the modification indices suggests further
refining of the measurement model.

References
Du Toit, M. & Du Toit, S.H.C. (2001).
Interactive LISREL: User’s Guide.
Lincolnwood, IL: Scientific Software International, Inc.
Jöreskog, K.G. & Sörbom, D. (1999).
LISREL 8: User’s Reference Guide.
Lincolnwood, IL: Scientific Software International, Inc.
Jöreskog, K.G. & Sörbom, D. (2006).
LISREL 8.80 for Windows [Computer Software].
Lincolnwood, IL: Scientific Software International, Inc.
Steenkamp, J.E.M. & Van Trijp, H.C.M (1991).
The use of LISREL in validating marketing constructs.
International Journal of Research in Marketing, 8, 283-299.

LISREL output files for measurement models 15